Pseudo-Single-Cell Spatial Proteomics of Human Pancreatic Islets

2,215 protein abundance maps at cellular resolution

Shane S. Kelly

⌄

This project presents a novel spatial proteomics workflow that maps protein expression at pseudo-single-cell resolution in human pancreatic islet tissue.

Combining nanopots mass spectrometry with immunofluorescence-guided cell segmentation, we resolve protein abundance across three major cell types—beta (β), alpha (α), and acinar—within a single tissue section.

Cell nuclei were identified with CellPose 2.0 deep learning segmentation, boundaries refined in FIJI, and cell types assigned by INS/GCG/DAPI immunofluorescence channel intensities. Single-cell RNA-seq reference data (Azimuth) was used to deconvolve pixel-level proteomics into cell-type-adjusted protein maps.

H&E stained pancreatic islet tissue at 10X magnification
H&E stained human pancreatic islet tissue (10X)
2,215Proteins Mapped
3Cell Types Resolved
6Pipeline Stages

From Tissue to Protein Maps

Drag the slider to compare cell type classification with protein expression at single-cell resolution.

DAPI Fluorescence vs. CellPose Nuclear Segmentation
DAPI nuclear fluorescence CellPose 2.0 nuclear segmentation mask
DAPI Nuclear Stain CellPose 2.0 Segmentation Mask
DAPI Fluorescence vs. Cell Boundary Overlay
DAPI nuclear fluorescence Cell boundaries segmentation overlay
DAPI Nuclear Stain Cell Boundary Segmentation
Cell Boundary Overlay vs. ROI Color Mask
Cell boundaries segmentation overlay 8-bit color ROI mask
Cell Boundaries ROI Color Mask
Cell Type Assignment vs. Insulin Expression
Cell type assignment map Insulin protein expression map
■ Acinar  ■ Alpha  ■ Beta INS — Low to High
Cell Type Assignment vs. Glucagon Expression
Cell type assignment map Glucagon protein expression map
■ Acinar  ■ Alpha  ■ Beta GCG — Low to High

Analysis Pipeline

A six-stage workflow from tissue imaging to spatially resolved protein maps.

1

3D Spatial Proteomics

Giotto
2D spatial plot
➔
2

Cell Segmentation

CellPose 2.0
DAPI nuclear segmentation
➔
3

Cell Type Assignment

IF Classification
Cell type classification
➔
4

Pixel Mapping

FIJI / ImageJ
Pixel assignment
➔
5

MS + RNA-Seq

Azimuth / Seurat
Azimuth pancreas reference UMAP
➔
6

Protein Maps

Cell-Type Adjusted
INS protein map

Marker Protein Maps

Validation of spatial protein maps against known cell type markers.

Islet protein markers: Secretogranin-2, Insulin, Glucagon
Islet Cell Markers. Secretogranin-2 (SCG2) shows broad islet expression. Insulin (INS) localizes to beta cells in the islet core. Glucagon (GCG) maps to alpha cells at the islet periphery. Navy = low, yellow = high relative intensity.
Acinar protein markers: REG1A, REG1B, CPA1
Acinar Cell Markers. REG1A, REG1B, and CPA1 show strong expression in the surrounding acinar tissue with minimal signal in the islet region, confirming successful cell-type-adjusted deconvolution.

Methods

Tissue Preparation & Imaging

Human pancreatic tissue sections were imaged with multiplex immunofluorescence using DAPI (nuclear), INS (insulin/beta cells), and GCG (glucagon/alpha cells) channels at 10X magnification.

Cell Segmentation

Nuclear boundaries were detected using CellPose 2.0 deep learning segmentation on the DAPI channel. Cell boundaries were refined and ROIs extracted in FIJI/ImageJ using custom macros.

Cell Type Assignment

Each cell was classified as alpha, beta, or acinar based on RGB intensity thresholds from the three immunofluorescence channels.

Spatial Mass Spectrometry Integration

Cells were linked to nanopots mass spectrometry pixels via 8-bit color histogram matching. Protein abundances (log2-transformed) were merged with spatial cell annotations.

RNA-Seq Deconvolution

Cell-type-specific protein expression ratios were calculated from the Azimuth human pancreas single-cell RNA-seq reference dataset (Seurat). These ratios deconvolve pixel-level proteomics into cell-type-adjusted abundance maps.

Protein Map Generation

For each of 2,215 proteins passing quality filters, cell-type-adjusted relative abundance was calculated and visualized as spatial maps using ggplot2 with sf polygon geometries.